The study sample consisted of 44,957 person-years, of whom 45% were men; 80% were white; 45% were 75 years and older; 73% resided in a metropolitan area; 58% had incomes below 200% federal poverty level; 67% had no college education; and 18% had dual Medicare/Medicaid coverage.
Among veterans, of 11,759 person-years, 2,516 were predominant-VHA users, 452 had some VHA use, and 8,791 were non-VHA users. Predominant VHA users were less likely to have high income, prescription drug coverage and good health status (Table ). For example, only 12% of predominant VHA users were in the high income group whereas 22% of veterans with no VHA use were in the high income group. Similarly, 57% of predominant VHA users lacked prescription drug coverage, whereas 21% of veterans with no VHA use lacked the same. Higher rates of diabetes (27% versus 19%), heart disease (46% versus 36%), and hypertension (71% versus 57%) were observed in veterans with predominant VHA use as opposed to their veteran counterparts with no VHA use.
Among all Medicare beneficiaries, 5% (N
2,311) had any ACSH. Among those who had inpatient use, 24.3% had any ACSH; 10.1% had any acute ACSH; 15.8% had any chronic ACSH. Of the thirteen ACSH measured, the three most common hospitalizations were for congestive heart failure (7.3%), bacterial pneumonia (6.7%) and COPD (4.8%) (data not shown in tabular form).
Significant subgroup differences in rates of ACSH by race/ethnicity, age, marital status, metropolitan status and geographic region, education, income, insurance status and all health measures were noted (Table ). When examined by dual use, the weighted average ACSH rate for veterans with predominant VHA use was 4.9%; 3.7% for some VHA use; and 4.5 % for veterans with no VHA use. This difference was not statistically significant.
Weighted Percent with Any ACSH and Adjusted Odds Ratios and 95% Confidence Intervals from Logistic Regression on Any ACSH Medicare Current Beneficiary Survey, 2001–2005
Table also presents the adjusted odds ratios (AOR) and 95% confidence intervals (CI) from logistic regressions on presence of any ACSH. Compared to veterans with no VHA use, the risk of any ACSH was lower among veterans with predominant VHA use [AOR
0.84, 95% CI
0.64, 1.11]. However, the difference was not statistically significant. The risk of any ACSH was lower for some VHA use when compared to veterans with no VHA use; however the lower risk was also not statistically significant [AOR
0.71,95%CI: 0.42 , 1.19].
To ensure the robustness of the relationship between any ACSH and dual use, we analyzed the data by subgroups. These subgroups included only men and those with cardio-metabolic conditions namely, diabetes, heart disease, or hypertension. In both these sets of analyses, the findings remained consistent with the primary analyses.
Controlling for Selection Bias for Dual Use
As seen from Table , dual users may be a select subgroup based on poverty status, prescription drug coverage, and health status. Dual users may also differ from veterans with no VHA use by omitted variables such as VHA priority status (service-related disabilities and low-income) and unobserved variables such as veteran preferences for choosing Medicare or VHA. For example, prior literature suggests that veterans with higher VA priority status, better health status, and living near a VA facility were less likely to be dual users.12
Similarly, unobserved factors such as perceived fragmentation of care and patient satisfaction may affect dual use.21
Therefore, we controlled for the selection and reverse causality by using an instrumental variable approach. We used distance between the resident county and nearest VHA facility as an instrument in the selection equation that predicted presence/absence of dual use. The model on ACSH included all independent variables mentioned in the measures section. This analysis was restricted to veterans and was conducted in STATA 11.22
The instrumental variable approach revealed that even after controlling for selection into dual use, there was no statistically significant relationship between dual use and any ACSH. Dual users were less likely to have any ACSH (parameter estimate
0.20); but the relationship was not statistically significant. The relationship remained consistent when testing various specifications of distance (i.e. driving distance, straight line distance, driving time, and categories of driving distance and driving time).